Artificial Neural Network Analysis on Suicide and Self-Harm History of Taiwanese Soldiers

There are two potential risk factors of suicide commit have been proved nowadays, namely past history of serious suicide idea and of deliberate self- harm. Our study attempted to use artificial neural network (ANN) to predict those past histories from other eight current ordinary factors, such as age, years of education, religion, family status, past psychiatry history, family psychiatry history, anxiety status and depression status. We collected 225 self-administrated results from three different group ROC soldiers, including, troops in Taiwan, troops in isolated islands and psychiatry inpatients from September 2005 to April 2006. Randomly selected 25% of each group were the testing group and the rests were the training group, which trained by radial basis function (RBF) models. As the results, our trained model showed 81.8% as sensitivity and 85.7% as specificity in detecting past suicide idea history of testing group, meanwhile, 75.0% as sensitivity and 75.6% as specificity in detecting past self-harm history. Our study found that by using eight current general factors, RBF neural network models showed acceptable performance in detection of past suicide idea history as well as past self-harm history.